1 // Auto-generated file. Do not edit!
2 // Template: src/f32-vbinary/vop-scalar.c.in
3 // Generator: tools/xngen
4 //
5 // Copyright 2019 Google LLC
6 //
7 // This source code is licensed under the BSD-style license found in the
8 // LICENSE file in the root directory of this source tree.
9
10 #include <assert.h>
11
12 #include <xnnpack/common.h>
13 #include <xnnpack/math.h>
14 #include <xnnpack/vbinary.h>
15
16
xnn_f32_vadd_minmax_ukernel__wasm_x8(size_t n,const float * a,const float * b,float * y,const union xnn_f32_minmax_params params[restrict XNN_MIN_ELEMENTS (1)])17 void xnn_f32_vadd_minmax_ukernel__wasm_x8(
18 size_t n,
19 const float* a,
20 const float* b,
21 float* y,
22 const union xnn_f32_minmax_params params[restrict XNN_MIN_ELEMENTS(1)])
23 {
24 assert(n != 0);
25 assert(n % sizeof(float) == 0);
26 assert(a != NULL);
27 assert(b != NULL);
28 assert(y != NULL);
29
30 const float vy_min = params->scalar.min;
31 const float vy_max = params->scalar.max;
32
33 for (; n >= 8 * sizeof(float); n -= 8 * sizeof(float)) {
34 const float va0 = a[0];
35 const float va1 = a[1];
36 const float va2 = a[2];
37 const float va3 = a[3];
38 const float va4 = a[4];
39 const float va5 = a[5];
40 const float va6 = a[6];
41 const float va7 = a[7];
42 a += 8;
43
44 const float vb0 = b[0];
45 const float vb1 = b[1];
46 const float vb2 = b[2];
47 const float vb3 = b[3];
48 const float vb4 = b[4];
49 const float vb5 = b[5];
50 const float vb6 = b[6];
51 const float vb7 = b[7];
52 b += 8;
53
54 float vy0 = va0 + vb0;
55 float vy1 = va1 + vb1;
56 float vy2 = va2 + vb2;
57 float vy3 = va3 + vb3;
58 float vy4 = va4 + vb4;
59 float vy5 = va5 + vb5;
60 float vy6 = va6 + vb6;
61 float vy7 = va7 + vb7;
62
63
64 vy0 = __builtin_wasm_max_f32(vy0, vy_min);
65 vy1 = __builtin_wasm_max_f32(vy1, vy_min);
66 vy2 = __builtin_wasm_max_f32(vy2, vy_min);
67 vy3 = __builtin_wasm_max_f32(vy3, vy_min);
68 vy4 = __builtin_wasm_max_f32(vy4, vy_min);
69 vy5 = __builtin_wasm_max_f32(vy5, vy_min);
70 vy6 = __builtin_wasm_max_f32(vy6, vy_min);
71 vy7 = __builtin_wasm_max_f32(vy7, vy_min);
72
73 vy0 = __builtin_wasm_min_f32(vy0, vy_max);
74 vy1 = __builtin_wasm_min_f32(vy1, vy_max);
75 vy2 = __builtin_wasm_min_f32(vy2, vy_max);
76 vy3 = __builtin_wasm_min_f32(vy3, vy_max);
77 vy4 = __builtin_wasm_min_f32(vy4, vy_max);
78 vy5 = __builtin_wasm_min_f32(vy5, vy_max);
79 vy6 = __builtin_wasm_min_f32(vy6, vy_max);
80 vy7 = __builtin_wasm_min_f32(vy7, vy_max);
81
82 y[0] = vy0;
83 y[1] = vy1;
84 y[2] = vy2;
85 y[3] = vy3;
86 y[4] = vy4;
87 y[5] = vy5;
88 y[6] = vy6;
89 y[7] = vy7;
90 y += 8;
91 }
92 if XNN_UNLIKELY(n != 0) {
93 do {
94 const float va = *a++;
95 const float vb = *b++;
96 float vy = va + vb;
97 vy = __builtin_wasm_max_f32(vy, vy_min);
98 vy = __builtin_wasm_min_f32(vy, vy_max);
99 *y++ = vy;
100 n -= sizeof(float);
101 } while (n != 0);
102 }
103 }
104